Derived dependent atoms not updating during multiple sets

[jsheely]

Summary

I am seeing a change from 2.3.1 to 2.4.1 where atoms that are updated that have derived values using that atom are not updated

I commonly use backing fields for my atoms with custom setters that set the backing field and may set multiple other atoms.

The change between the versions is that when doing a set() on the backing field and then another set() for another atom() that uses that backing field. It is fine. But if that atom() uses another atom that uses the backing field it has not been updated yet.

Here is my quick psuedo code of the issue

const testingArray = [];
const _selectedItemIdAtom = atom<string>("")
const selectedItemIdAtom = atom(
    (get) => get(_selectedItemIdAtom),
    (get, set, id: string) => {
        set(_selectedItemIdAtom, id);
        if (id === '')
            set(stateAtom, null)
    }
)

const selectedItem = atom((get)=> {
    const id = get(selectedItemIdAtom);
    return testingArray.find(item => item.id === id);
})

const _stateAtom = atom({});
const stateAtom = atom((get) => { }, (get, set) => { 
    const item = get(selectedItem); 
    const itemId = get(_selectedItemIdAtom)
    console.log(itemId, item?.id) // itemId !== item?.id in 2.4.1. Correct in 2.3.1
    if (item) {
        set(_stateAtom, item.state);
    }
});

Link to reproduction

https://codesandbox.io/s/jotai-2-4-1-regression-kmclx3?file=/src/App.tsx

[dai-shi]

Thanks for reporting.
I hope you can create a small reproduction with https://csb.jotai.org.

Meanwhile, if this is a regression from 2.3.1, can you try and see which causes it?

• 2.4.0
• fix(vanilla): should only mount dependencies if the atom is mounted #2078
• fix(vanilla): should update with conditional dependencies #2086

You can install the codesandbox build in those PRs.

[jsheely]

Attached a repo. Changing from 2.4.1 to 2.3.1 in the package.json will change the behavior of this button

(2.4.1) Missmatch 1 undefined
(2.3.1) Missmatch 1 1
Console.log

I believe that it might be related to the work on #2086 which was my initial hunch. I will try and test them individually

[jsheely]

Confirmed that it is #2086 that introduces the behavior change

https://codesandbox.io/s/react-typescript-forked-57q86k

[dai-shi]

Thanks. Is it possible for you to try it with Jotai v1.13.1? #2086 was to fix a regression from v1.

[jsheely]

1.13.1 has the same behavior as 2.3.1 where the values match.

[dai-shi]

Thanks. Yeah, this seems a regression with fixing another regression from v1...

[dai-shi]

#2111 fixes it: https://codesandbox.io/s/jotai-2-4-1-regression-95584g

[jsheely]

Confirmed #2111 resolves issue. Nice work on the PR. That looks like a non trivial update.